Lesson 3 · 7 topics · 22 min · intermediate

Hybrid Search and Reranking

You will be able to

Combine retrievers in sequence or in parallel, and say what each pattern can lose.
Compute reciprocal rank fusion, and explain why it uses ranks rather than scores and what its constant k does.
Recognise when fusion hurts: one retriever's spurious match outvoting another's correct result.
Explain the difference between an embedding model (bi-encoder) and a cross-encoder reranker, and why rerankers only score a shortlist.
Size a reranker's candidate pool from where the right chunk usually lands.
Rerank by recency, and treat inclusion as more important than exact position in a RAG context.

Topics

Your review

Sign in to keep your quick-check answers. The ones you miss come back for review after 1, 3 and 7 days, until they stick.

Sources

AI Engineering · Chip Huyen, 2025
Chip Huyen. O'Reilly. Used: Ch. 6 "RAG and Agents": "Combining retrieval algorithms"; "Retrieval Optimization" > "Reranking".
Hands-On Large Language Models · Jay Alammar & Maarten Grootendorst, 2024
Jay Alammar & Maarten Grootendorst. O'Reilly. Used: Ch. 8 "Semantic Search and Retrieval-Augmented Generation": "Reranking" > "Reranking example", "How reranking models work".
Reciprocal Rank Fusion outperforms Condorcet and individual Rank Learning Methods · Gordon V. Cormack et al., 2009
Gordon V. Cormack, Charles L. A. Clarke & Stefan Büttcher. SIGIR 2009. Used: §1 and Table 1 (fetched 2026-10-23). Open source
Multi-Stage Document Ranking with BERT · Rodrigo Nogueira et al., 2019
Rodrigo Nogueira, Wei Yang, Kyunghyun Cho & Jimmy Lin. arXiv:1910.14424. Used: Abstract (fetched 2026-10-23). Open source
cross-encoder/ms-marco-MiniLM-L6-v2 (model card) · Sentence Transformers
Sentence Transformers. Used: Trained on the MS MARCO Passage Ranking task (fetched 2026-10-23). Open source

You will be able to

Combine retrievers in sequence or in parallel, and say what each pattern can lose.

Compute reciprocal rank fusion, and explain why it uses ranks rather than scores and what its constant k does.

Recognise when fusion hurts: one retriever's spurious match outvoting another's correct result.

Explain the difference between an embedding model (bi-encoder) and a cross-encoder reranker, and why rerankers only score a shortlist.

Size a reranker's candidate pool from where the right chunk usually lands.

Rerank by recency, and treat inclusion as more important than exact position in a RAG context.

Sources

AI Engineering · Chip Huyen, 2025

Chip Huyen. O'Reilly. Used: Ch. 6 "RAG and Agents": "Combining retrieval algorithms"; "Retrieval Optimization" > "Reranking".

Hands-On Large Language Models · Jay Alammar & Maarten Grootendorst, 2024

Jay Alammar & Maarten Grootendorst. O'Reilly. Used: Ch. 8 "Semantic Search and Retrieval-Augmented Generation": "Reranking" > "Reranking example", "How reranking models work".

Reciprocal Rank Fusion outperforms Condorcet and individual Rank Learning Methods · Gordon V. Cormack et al., 2009

Gordon V. Cormack, Charles L. A. Clarke & Stefan Büttcher. SIGIR 2009. Used: §1 and Table 1 (fetched 2026-10-23). Open source

Multi-Stage Document Ranking with BERT · Rodrigo Nogueira et al., 2019

Rodrigo Nogueira, Wei Yang, Kyunghyun Cho & Jimmy Lin. arXiv:1910.14424. Used: Abstract (fetched 2026-10-23). Open source

cross-encoder/ms-marco-MiniLM-L6-v2 (model card) · Sentence Transformers

Sentence Transformers. Used: Trained on the MS MARCO Passage Ranking task (fetched 2026-10-23). Open source